AI Investment Playbook Enters a New Phase as Chip Dominance Gives Way to Application-Driven Competition

GPU-centric investment cycle begins to shift
Data platforms and AI application services gain prominence
Rising preference for tangible assets reshapes investment priorities

Artificial intelligence (AI) investment strategies are reaching a new inflection point. The infrastructure race that had long been dominated by graphics processing units (GPUs) and data centers is broadening toward proprietary AI chips, industry-specific application platforms, and tangible assets that remain difficult for AI to replace. As Big Tech firms, biotechnology companies, and global private equity firms (PEFs) recalibrate their investment priorities, market attention is increasingly shifting away from AI technology itself toward sectors capable of delivering profitability and sustainable competitive advantages.

Big Tech Intensifies Push for In-House AI Chips

According to Bloomberg on July 1 (local time), OpenAI recently unveiled "Jalapeño," an AI chip co-developed with Broadcom. Optimized for inference, Jalapeño was designed to process the computational demands of ChatGPT and OpenAI's coding agent Codex more efficiently. The chip is specifically aimed at accelerating AI inference—the process through which AI models generate responses to user queries—while significantly reducing associated costs. Industry observers expect that while OpenAI will continue relying on Nvidia GPUs for large-scale pre-training of AI models, Jalapeño will substantially reduce the enormous expenses associated with inference.

OpenAI said Jalapeño is not a general-purpose accelerator derived from modifying or improving existing AI chips, but rather a processor designed from the ground up specifically for large language model (LLM) inference based on years of operating ChatGPT and Codex. The company completed the entire development process—from initial design to tape-out, the stage at which the chip design is transferred to semiconductor foundries—in just nine months, making it one of the fastest application-specific integrated circuit (ASIC) development cycles on record. Considering that designing a custom ASIC from scratch typically takes between 18 months and two years, the pace is regarded as exceptional. OpenAI attributed the shortened development timeline to its use of AI models throughout the design and optimization process.

Competitor Anthropic is also expected to begin developing its own AI chip in the near future. Reuters previously reported that surging computational demand this year has reached levels that are becoming increasingly difficult for the company to manage, prompting executives to review plans for proprietary AI silicon. Meanwhile, Google, Amazon, Microsoft (MS), and Meta—all of which are investing astronomical sums in AI infrastructure this year—have already introduced their own AI chips. Google has utilized its in-house Tensor Processing Units (TPUs), first developed in 2015, across its core services and cloud business, and has recently signed agreements to supply TPUs to external companies including Anthropic and Meta.

Amazon is pursuing plans to sell its internally developed Trainium training chips and Inferentia inference chips directly to outside customers without routing sales exclusively through Amazon Web Services (AWS). The strategy targets companies seeking to reduce dependence on Nvidia hardware. Since launching Trainium in 2020, Amazon has supplied the chips through AWS to customers including OpenAI, Anthropic, and Uber. Revenue commitments generated by the chips had reportedly surpassed $225 billion as of April this year.

Microsoft has also deployed its Maia 200 inference-optimized AI chips across major data centers and is discussing supplying them to Anthropic. Meta unveiled four versions of its proprietary Meta Training and Inference Accelerator (MTIA) chips in March. Explaining its diversified AI chip strategy—which combines externally sourced chips such as Nvidia GPUs with proprietary silicon depending on workloads—Meta noted that general-purpose chips are optimized for AI training, the industry's most demanding computing task, making them less cost-efficient for inference workloads. Accordingly, the company intends to deploy MTIA primarily for inference.

Biotech AI Investment Shifts Toward Data Platforms

The changing AI strategy extends well beyond semiconductors. The center of gravity in AI investment is also shifting rapidly. Over the past two years, capital largely flowed into infrastructure expansion, including AI model training and data center construction. More recently, however, investment has increasingly migrated toward commercializing AI through real-world services and industry applications. As the core of AI competition expands beyond computational capacity toward utilization and productivity, investment criteria are evolving accordingly.

The software industry is where this trend is most apparent. Companies are increasingly prioritizing investments in workflow automation, enterprise AI agents, and industry-specific AI solutions rather than competing to build general-purpose AI models. Whereas hardware investments require substantial capital expenditures (CAPEX), software businesses can achieve greater investment efficiency through operating expenditure (OPEX)-based subscription software-as-a-service (SaaS) models. As software companies that were previously left behind during the semiconductor rally begin demonstrating accelerating revenue growth and operating margins, investors are increasingly attracted by their superior capital efficiency.

The biotechnology sector is moving in the same direction. Major global pharmaceutical companies are now placing greater emphasis on platforms that accumulate and manage experimental data rather than on AI models themselves. Merck (MSD) strengthened its AI-driven drug discovery platform by signing joint development agreements worth a combined $1.23 billion with UK-based AI drug discovery firms BenevolentAI and Exscientia. Pfizer has also built its proprietary AI platform, VOX, and deployed it across 19 drug development programs. Sanofi is likewise accelerating the advancement of its R&D platform through partnerships with specialized AI companies.

Their investment strategy differs fundamentally from the previous race to develop algorithms. Instead, resources are being concentrated on standardizing experimental datasets, protein information, and clinical data accumulated throughout the research process so that AI systems can repeatedly learn from and validate them. Merck's recent research collaboration with Protilion reflects the same philosophy. Protilion's Prot-MaP platform generates massive protein datasets for AI model training while simultaneously evaluating binding affinity and manufacturing suitability across large numbers of drug candidates. Industry observers increasingly believe that competition over data platforms—rather than AI model performance alone—has become the decisive factor in improving drug development efficiency.

Private Equity Shifts Capital Toward AI-Resilient Assets

Private equity investment is also undergoing a strategic realignment. While capital continues flowing into AI-related assets, a growing share is now targeting industries that remain difficult for AI to replace. A representative example is the "HALO (Human Advantage, Asset-backed, Local, Offline)" strategy that has gained traction among global private equity firms. Leading asset managers including Blackstone, Brookfield, Kohlberg Kravis Roberts (KKR), and Apollo are increasing allocations not only to AI infrastructure such as data centers and power grids but also to sectors where human expertise and tangible assets remain decisive competitive advantages, including healthcare services, waste management, logistics facilities, education, essential consumer goods, and locally based service businesses.

Specifically, multiple private equity firms are reviewing bids for Volkswagen's medium- and large-displacement diesel engine division, while financial investors (FIs) are also competing to acquire UK aerospace components manufacturer Senior. Discussions surrounding the sale of TK Elevator are likewise progressing on the basis of an enterprise valuation of approximately $29.4 billion, underscoring the continued concentration of capital in infrastructure and industrial assets. Investment standards are evolving alongside these transactions. Market participants increasingly view the changing definition of what deserves a valuation premium as especially significant. Across global investment markets, tangible assets backed by visible demand are now commanding stronger investor confidence than high-growth narratives alone.

A similar trend is emerging in South Korea. Investment professionals say the so-called "K-Curve" phenomenon is becoming increasingly pronounced. Companies that rapidly improve productivity and profitability through AI adoption are attracting growing capital inflows, while businesses lacking both clear AI strategies and differentiated competitiveness are falling down investors' priority lists. Conversely, industries such as healthcare, defense, infrastructure, and premium consumer goods—where direct AI substitution remains limited—continue drawing steady investment. Collectively, the changes unfolding across semiconductors, software, biotechnology, and financial markets point in the same direction. While the early phase of AI investing centered on securing GPUs and building data centers, capital is now dispersing toward proprietary AI chips, industry-specific application platforms, and business models capable of maintaining competitive advantages throughout the AI era.